appleboy/CodeGPT

CodeGPT

A CLI tool written in Go that generates git commit messages or provides code review summaries using ChatGPT AI (gpt-5, gpt-4o, gpt-4 model). It also automatically installs a git prepare-commit-msg hook.

• 繁體中文介紹
• 繁體中文影片

English | 繁體中文 | 简体中文

Table of Contents

• CodeGPT
  • Table of Contents
  • Features
  • Installation
    • macOS
    • Windows
    • Install via Script or Binary
      • A. Automated Install (Recommended)
      • B. Manual Install (Advanced)
      • Configurable Environment Variables
    • From Source
    • Using VSCode Devcontainer
  • Configuration
    • Using API Key Helper for Dynamic Credentials
      • Setup API Key Helper
      • Configure Refresh Interval
      • Gemini-Specific API Key Helper
      • How It Works
      • Priority Order
    • How to Customize the Default Prompt Folder
    • How to Change to Azure OpenAI Service
    • Support for Gemini API Service
      • Configuration Options
      • Example: Gemini API (default backend)
      • Example: VertexAI Gemini
    • Support for Anthropic API Service
    • How to Change to Groq API Service
    • How to Change to Ollama API Service
    • How to Change to OpenRouter API Service
  • Usage
    • CLI Mode
  • Change Commit Message Template
    • Git Hook
      • Install
      • Uninstall
    • Code Review
  • Testing
  • Star History
  • Reference

Features

• Supports Azure OpenAI Service, OpenAI API, Gemini, Anthropic, Ollama, Groq, and OpenRouter.
• Adheres to the conventional commits specification.
• Integrates with Git prepare-commit-msg Hook, see the Git Hooks documentation.
• Allows customization of generated diffs by specifying the number of context lines (default: 3).
• Enables file exclusion patterns for git diff operations.
• Translates commit messages into other languages (supports en, zh-tw, or zh-cn).
• Supports SOCKS proxy or custom network HTTP proxy.
• Generates brief code reviews.
• Supports streaming output for real-time token display.
• Allows customization of prompt templates and variables.

Installation

macOS

Install via Homebrew:

brew tap appleboy/tap
brew install codegpt

Windows

Install via Chocolatey:

choco install codegpt

Install via Script or Binary

For most systems, you can use an install script or manually download a pre-compiled binary.

A. Automated Install (Recommended)

Run the install script to automatically download and set up the latest release:

bash < <(curl -sSL https://raw.githubusercontent.com/appleboy/CodeGPT/main/install.sh)

Or download and run manually:

curl -LO https://raw.githubusercontent.com/appleboy/CodeGPT/main/install.sh
chmod +x install.sh
./install.sh

B. Manual Install (Advanced)

1. Download the latest pre-compiled binary from the release page.

2. Change permissions:

   chmod 755 codegpt
   

3. Move the binary to your system bin directory:

   mv codegpt /usr/local/bin/
   

4. Confirm installation:

   codegpt version
   

Configurable Environment Variables

Example usage:

# Install with insecure mode enabled (ignoring curl SSL verification):
INSECURE=1 ./install.sh

# Install a specific version to a custom directory:
VERSION=1.1.0 INSTALL_DIR=/opt/codegpt ./install.sh

The script will:

1. Detect your OS and architecture
2. Download the latest binary
3. Add the binary to your PATH

From Source

Install from source code:

go install github.com/appleboy/CodeGPT/cmd/codegpt@latest

Using VSCode Devcontainer

Add the feature to your devcontainer.json:

"features": {
    "ghcr.io/kvokka/features/codegpt:1": {}
}

Configuration

First, create your OpenAI API Key. The OpenAI Platform allows you to generate a new API Key.

Set the environment variable OPENAI_API_KEY:

export OPENAI_API_KEY=sk-xxxxxxx

Alternatively, store your API key in a custom config file:

codegpt config set openai.api_key sk-xxxxxxx

This will create a .codegpt.yaml file in your home directory ($HOME/.config/codegpt/.codegpt.yaml). The following options are available:

Using API Key Helper for Dynamic Credentials

Instead of storing your API key directly in the config file, you can use a shell command to dynamically retrieve it from a password manager or secret service. This is especially useful for:

• Fetching keys from password managers (1Password, Bitwarden, etc.)
• Using cloud secret services (AWS Secrets Manager, Google Secret Manager, etc.)
• Implementing token rotation and short-lived credentials
• Enhancing security by not storing keys in plain text

Setup API Key Helper

Configure a shell command to retrieve your API key:

# Using 1Password CLI
codegpt config set openai.api_key_helper "op read op://vault/openai/api_key"

# Using AWS Secrets Manager
codegpt config set openai.api_key_helper "aws secretsmanager get-secret-value --secret-id openai-key --query SecretString --output text"

# Using Google Cloud Secret Manager
codegpt config set openai.api_key_helper "gcloud secrets versions access latest --secret=openai-api-key"

# Using environment variable
codegpt config set openai.api_key_helper "echo \$MY_OPENAI_KEY"

# Custom script
codegpt config set openai.api_key_helper "/path/to/get-api-key.sh"

Configure Refresh Interval

By default, the API key is cached for 15 minutes (900 seconds) to avoid excessive calls to your secret service:

# Set refresh interval to 5 minutes
codegpt config set openai.api_key_helper_refresh_interval 300

# Set refresh interval to 30 minutes
codegpt config set openai.api_key_helper_refresh_interval 1800

# Disable caching (fetch key every time)
codegpt config set openai.api_key_helper_refresh_interval 0

Gemini-Specific API Key Helper

For Gemini provider, you can set a separate helper:

codegpt config set gemini.api_key_helper "gcloud secrets versions access latest --secret=gemini-key"
codegpt config set gemini.api_key_helper_refresh_interval 600

How It Works

1. First execution: CodeGPT runs your helper command and caches the API key in ~/.config/codegpt/.cache/ with restrictive permissions (0600)
2. Subsequent executions: Within the refresh interval, CodeGPT uses the cached key
3. After expiration: CodeGPT automatically re-runs the helper command and updates the cache
4. Security: Cache files are stored with owner-only read/write permissions

Priority Order

When multiple API key sources are configured, CodeGPT uses this priority:

1. openai.api_key_helper (if configured)
2. openai.api_key (static config)
3. OPENAI_API_KEY environment variable

How to Customize the Default Prompt Folder

The default prompt folder is located at $HOME/.config/codegpt/prompt. You can change this to another directory by executing:

codegpt config set prompt.folder /path/to/your/prompt

To load the prompt files from the custom folder, run:

codegpt prompt --load

Upon execution, you will see messages similar to the following:

save code_review_file_diff.tmpl to /Users/xxxxx/.config/codegpt/prompt/code_review_file_diff.tmpl
save summarize_file_diff.tmpl to /Users/xxxxx/.config/codegpt/prompt/summarize_file_diff.tmpl
save summarize_title.tmpl to /Users/xxxxx/.config/codegpt/prompt/summarize_title.tmpl
save conventional_commit.tmpl to /Users/xxxxx/.config/codegpt/prompt/conventional_commit.tmpl

• code_review_file_diff.tmpl
• summarize_file_diff.tmpl
• summarize_title.tmpl
• conventional_commit.tmpl

How to Change to Azure OpenAI Service

Get the API key, Endpoint, and Model deployments list from the Azure Resource Management Portal on the left menu.

Update your config file:

codegpt config set openai.provider azure
codegpt config set openai.base_url https://xxxxxxxxx.openai.azure.com/
codegpt config set openai.api_key xxxxxxxxxxxxxxxx
codegpt config set openai.model xxxxx-gpt-4o

Support for Gemini API Service

You can use the Gemini API or VertexAI Gemini service. See the Gemini API documentation and VertexAI documentation.
Update the following parameters in your config file.

• Please create an API key from the Gemini API page (for BackendGeminiAPI) or from VertexAI API Key (for BackendVertexAI).

Configuration Options

Example: Gemini API (default backend)

codegpt config set openai.provider gemini
codegpt config set openai.model gemini-2.0-flash
codegpt config set gemini.api_key xxxxxxx
# gemini.backend defaults to BackendGeminiAPI, so you can omit it

Example: VertexAI Gemini

codegpt config set openai.provider gemini
codegpt config set openai.model gemini-2.0-flash
codegpt config set gemini.backend BackendVertexAI
codegpt config set gemini.project_id my-gcp-project
codegpt config set gemini.location us-central1

flowchart TD
    User([User])
    subgraph CodeGPT
      GeminiClient([Gemini Provider])
    end
    subgraph Google
      GeminiAPI([Gemini API])
      VertexAI([VertexAI Gemini])
    end

    User -->|Completion / GetSummaryPrefix| GeminiClient

    GeminiClient -- BackendGeminiAPI --> GeminiAPI
    GeminiAPI -- Response (text, usage) --> GeminiClient

    GeminiClient -- BackendVertexAI --> VertexAI
    VertexAI -- Response (text, usage) --> GeminiClient

    GeminiClient --> User

Support for Anthropic API Service

Build with the Anthropic API, you can see the Anthropic API documentation. Update the provider and api_key in your config file. Please create an API key from the Anthropic API page.

codegpt config set openai.provider anthropic
codegpt config set openai.api_key xxxxxxx
codegpt config set openai.model claude-3-5-sonnet-20241022

See the model list from the Anthropic API documentation.

How to Change to Groq API Service

Get the API key from the Groq API Service by visiting the Groq API Service. Update the base_url and api_key in your config file.

codegpt config set openai.provider openai
codegpt config set openai.base_url https://api.groq.com/openai/v1
codegpt config set openai.api_key gsk_xxxxxxxxxxxxxx
codegpt config set openai.model llama3-8b-8192

GroqCloud currently supports the following models:

1. Production Models
2. Preview Models

How to Change to Ollama API Service

We can use the Llama3 model from the Ollama API Service; please visit the Ollama API Models documentation. Update the base_url in your config file.

# pull llama3 8b model
ollama pull llama3
ollama cp llama3 gpt-4o

Try to use the Ollama API Service:

curl http://localhost:11434/v1/chat/completions \
  -H "Content-Type: application/json" \
  -d '{
    "model": "gpt-4o",
    "messages": [
      {
        "role": "user",
        "content": "Hello!"
      }
    ]
  }'

Update the base_url in your config file. You don't need to set the api_key in the config file.

codegpt config set openai.base_url http://localhost:11434/v1

How to Change to OpenRouter API Service

You can see the supported models list, model usage can be paid by users, developers, or both, and may shift in availability. You can also fetch models, prices, and limits via API.

The following example uses the free model name: meta-llama/llama-3-8b-instruct:free

codegpt config set openai.provider openai
codegpt config set openai.base_url https://openrouter.ai/api/v1
codegpt config set openai.api_key sk-or-v1-xxxxxxxxxxxxxxxx
codegpt config set openai.model google/gemini-3-flash-preview

To include your app in openrouter.ai rankings and show it in rankings on openrouter.ai, you can set the openai.headers in your config file:

codegpt config set openai.headers "HTTP-Referer=https://github.com/appleboy/CodeGPT X-Title=CodeGPT"

• HTTP-Referer: Optional, for including your app in openrouter.ai rankings.
• X-Title: Optional, for showing in rankings on openrouter.ai.

Usage

There are two methods for generating a commit message using the codegpt command: CLI mode and Git Hook.

CLI Mode

You can call codegpt directly to generate a commit message for your staged changes:

git add <files...>
codegpt commit --preview

The commit message is shown below:

Summarize the commit message using the gpt-4o model
We are trying to summarize a git diff
We are trying to summarize a title for the pull request
================Commit Summary====================

feat(cmd): Add preview flag and remove disableCommit flag in commit command and template file.

- Add a `preview` flag to the `commit` command
- Remove the `disableCommit` flag from the `prepare-commit-msg` template file

==================================================
Write the commit message to .git/COMMIT_EDITMSG file

Or translate all git commit messages into a different language (Traditional Chinese, Simplified Chinese, or Japanese):

codegpt commit --lang zh-tw --preview

Consider the following outcome:

Summarize the commit message using the gpt-4o model
We are trying to summarize a git diff
We are trying to summarize a title for the pull request
We are trying to translate a git commit message to Traditional Chinese language
================Commit Summary====================

功能：重構 codegpt commit 命令標記

- 將「codegpt commit」命令新增「預覽」標記
- 從「codegpt commit」命令中移除「--disableCommit」標記

==================================================
Write the commit message to .git/COMMIT_EDITMSG file

You can replace the tip of the current branch by creating a new commit. Just use the --amend flag:

codegpt commit --amend

Enable streaming output to see tokens as they arrive in real-time, rather than waiting for the full response:

codegpt commit --stream

Or enable it globally via config:

codegpt config set openai.stream true

The --stream flag also works with the review command:

codegpt review --stream

Agent Skill Integration

CodeGPT provides agent skills such as /commit-message that generate conventional commit messages from your staged changes. In Claude Code, simply type /commit-message to analyze your staged diff and produce a well-formatted commit message with the appropriate prefix, scope, and summary.

The skill definitions are maintained in a separate repository: appleboy/skills.

Change Commit Message Template

The default commit message template is as follows. The summarize_prefix variable outputs the conventional commit prefix with scope in prefix(scope) format (e.g., feat(cmd), fix(git), docs(readme)):

{{ .summarize_prefix }}: {{ .summarize_title }}

{{ .summarize_message }}

Change the format with a template string using the --template_string parameter:

codegpt commit --preview --template_string \
  "[{{ .summarize_prefix }}]: {{ .summarize_title }}"

Change the format with a template file using the --template_file parameter:

codegpt commit --preview --template_file your_file_path

Add a custom variable to the git commit message template:

{{ .summarize_prefix }}: {{ .summarize_title }}

{{ .summarize_message }}

{{ if .JIRA_URL }}{{ .JIRA_URL }}{{ end }}

Add a custom variable to the git commit message template using the --template_vars parameter:

codegpt commit --preview --template_file your_file_path --template_vars \
  JIRA_URL=https://jira.example.com/ABC-123

Load a custom variable from a file using the --template_vars_file parameter:

codegpt commit --preview --template_file your_file_path --template_vars_file your_file_path

See the template_vars_file format as follows:

JIRA_URL=https://jira.example.com/ABC-123

Git Hook

You can also use the prepare-commit-msg hook to integrate codegpt with Git. This allows you to use Git normally and edit the commit message before committing.

Install

To install the hook in the Git repository:

codegpt hook install

Uninstall

To remove the hook from the Git repository:

codegpt hook uninstall

Stage your files and commit after installation:

git add <files...>
git commit

codegpt will generate the commit message for you and pass it back to Git. Git will open it with the configured editor for you to review/edit it. Then, to commit, save and close the editor!

$ git commit
Summarize the commit message using the gpt-4o model
We are trying to summarize a git diff
We are trying to summarize a title for the pull request
================Commit Summary====================

Improve user experience and documentation for OpenAI tools

- Add download links for pre-compiled binaries
- Include instructions for setting up OpenAI API key
- Add a CLI mode for generating commit messages
- Provide references for OpenAI Chat completions and ChatGPT/Whisper APIs

==================================================
Write the commit message to .git/COMMIT_EDITMSG file
[main 6a9e879] Improve user experience and documentation for OpenAI tools
 1 file changed, 56 insertions(+)

Code Review

You can use codegpt to generate a code review message for your staged changes:

codegpt review

Or translate all code review messages into a different language (Traditional Chinese, Simplified Chinese, or Japanese):

codegpt review --lang zh-tw

See the following result:

Code review your changes using gpt-4o model
We are trying to review code changes
PromptTokens: 1021, CompletionTokens: 200, TotalTokens: 1221
We are trying to translate core review to Traditional Chinese language
PromptTokens: 287, CompletionTokens: 199, TotalTokens: 486
================Review Summary====================

總體而言，此程式碼修補似乎在增加 Review 指令的功能，允許指定輸出語言並在必要時進行翻譯。以下是需要考慮的潛在問題：

- 輸出語言沒有進行輸入驗證。如果指定了無效的語言代碼，程式可能會崩潰或產生意外結果。
- 此使用的翻譯 API 未指定，因此不清楚是否存在任何安全漏洞。
- 無法處理翻譯 API 調用的錯誤。如果翻譯服

==================================================

Example php code review:

<?php
if( isset( $_POST[ 'Submit' ]  ) ) {
  // Get input
  $target = $_REQUEST[ 'ip' ];
  // Determine OS and execute the ping command.
  if( stristr( php_uname( 's' ), 'Windows NT' ) ) {
    // Windows
    $cmd = shell_exec( 'ping  ' . $target );
  }
  else {
    // *nix
    $cmd = shell_exec( 'ping  -c 4 ' . $target );
  }
  // Feedback for the end user
  $html .= "<pre>{$cmd}</pre>";
}
?>

code review result:

================Review Summary====================

Code review:

1. Security: The code is vulnerable to command injection attacks as the user input is directly used in the shell_exec() function. An attacker can potentially execute malicious commands on the server by injecting them into the 'ip' parameter.
2. Error handling: There is no error handling in the code. If the ping command fails, the error message is not displayed to the user.
3. Input validation: There is no input validation for the 'ip' parameter. It should be validated to ensure that it is a valid IP address or domain name.
4. Cross-platform issues: The code assumes that the server is either running Windows or *nix operating systems. It may not work correctly on other platforms.

Suggestions for improvement:

1. Use escapeshellarg() function to sanitize the user input before passing it to shell_exec() function to prevent command injection.
2. Implement error handling to display error messages to the user if the ping command fails.
3. Use a regular expression to validate the 'ip' parameter to ensure that it is a valid IP address or domain name.
4. Use a more robust method to determine the operating system, such as the PHP_OS constant, which can detect a wider range of operating systems.

==================================================

Testing

Run the following command to test the code:

make test

Star History

Reference

• OpenAI Chat completions documentation.
• Introducing ChatGPT and Whisper APIs